diff options
| author | Joseph Huber <huberjn@outlook.com> | 2025-10-06 12:49:44 -0500 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-10-06 12:49:44 -0500 |
| commit | 8763812b4c60a702094d03c45d3a9db4478ca331 (patch) | |
| tree | 42c973cd5a1301a3685407c5fc45d3447a179f63 /offload | |
| parent | 839b91c2294b4aeb5598309f90afa241ace5acef (diff) | |
[Offload] Remove check on kernel argument sizes (#162121)
Summary:
This check is unnecessarily restrictive and currently incorrectly fires
for any size less than eight bytes. Just remove it, we do sanity checks
elsewhere and at some point need to trust the ABI.
Diffstat (limited to 'offload')
4 files changed, 6 insertions, 5 deletions
diff --git a/offload/plugins-nextgen/amdgpu/src/rtl.cpp b/offload/plugins-nextgen/amdgpu/src/rtl.cpp index f73fa0475a3a..8d2f9755e035 100644 --- a/offload/plugins-nextgen/amdgpu/src/rtl.cpp +++ b/offload/plugins-nextgen/amdgpu/src/rtl.cpp @@ -3687,11 +3687,6 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice, KernelArgsTy &KernelArgs, KernelLaunchParamsTy LaunchParams, AsyncInfoWrapperTy &AsyncInfoWrapper) const { - if (ArgsSize != LaunchParams.Size && - ArgsSize > LaunchParams.Size + getImplicitArgsSize()) - return Plugin::error(ErrorCode::INVALID_ARGUMENT, - "invalid kernel arguments size"); - AMDGPUPluginTy &AMDGPUPlugin = static_cast<AMDGPUPluginTy &>(GenericDevice.Plugin); AMDHostDeviceTy &HostDevice = AMDGPUPlugin.getHostDevice(); diff --git a/offload/unittests/OffloadAPI/device_code/CMakeLists.txt b/offload/unittests/OffloadAPI/device_code/CMakeLists.txt index 50e430597e64..1a042e1b3831 100644 --- a/offload/unittests/OffloadAPI/device_code/CMakeLists.txt +++ b/offload/unittests/OffloadAPI/device_code/CMakeLists.txt @@ -2,6 +2,7 @@ add_offload_test_device_code(foo.cpp foo) add_offload_test_device_code(bar.cpp bar) # Compile with optimizations to eliminate AMDGPU implicit arguments. add_offload_test_device_code(noargs.cpp noargs -O3) +add_offload_test_device_code(byte.cpp byte) add_offload_test_device_code(localmem.cpp localmem) add_offload_test_device_code(localmem_reduction.cpp localmem_reduction) add_offload_test_device_code(localmem_static.cpp localmem_static) @@ -14,6 +15,7 @@ add_custom_target(offload_device_binaries DEPENDS foo.bin bar.bin noargs.bin + byte.bin localmem.bin localmem_reduction.bin localmem_static.bin diff --git a/offload/unittests/OffloadAPI/device_code/byte.cpp b/offload/unittests/OffloadAPI/device_code/byte.cpp new file mode 100644 index 000000000000..779d120fefca --- /dev/null +++ b/offload/unittests/OffloadAPI/device_code/byte.cpp @@ -0,0 +1,3 @@ +#include <gpuintrin.h> + +extern "C" __gpu_kernel void byte(unsigned char c) { (void)c; } diff --git a/offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp b/offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp index 1dac8c50271b..c9eca36a4d44 100644 --- a/offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp +++ b/offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp @@ -55,6 +55,7 @@ struct LaunchSingleKernelTestBase : LaunchKernelTestBase { KERNEL_TEST(Foo, foo) KERNEL_TEST(NoArgs, noargs) +KERNEL_TEST(Byte, byte) KERNEL_TEST(LocalMem, localmem) KERNEL_TEST(LocalMemReduction, localmem_reduction) KERNEL_TEST(LocalMemStatic, localmem_static) |